Confounding by comorbidities is of concern in many epidemiological studies. To take this into account a common strategy is to calculate each participant's Charlson Comorbidity Index (CCI) and use this for adjustment in regression analyses. Various CCI adjustment strategies are possible, and it is unclear, which is preferable. In this simulation study, we compared common adjustment strategies in Cox regression analyses to determine to which degree they mitigate confounding and conservative bias caused by missing adjustment for independent predictors. We found that adjustment for each comorbidity as separate dichotomous covariate is the preferable adjustment strategy in samples of sufficient size as this mitigates both bias sources to the largest degree. If this is impractical in smaller studies adjustment for CCI split into multiple categories is preferable. In conclusion, the choice of CCI adjustment strategy impacts mitigation of bias in this simulation study, and suboptimal adjustment strategies can cause an observable bias, although of quite limited magnitude of only a few percent in this simulation example. Researcher should be careful when deciding on the adjustment strategies applied to ensure that the desired mitigation of bias sources is achieved.
Keywords: Charlson Comorbidity Index; Confounder adjustment; Epidemiology; Regression models; Simulation.
© 2021. Springer Nature B.V.